What Thaler v. Perlmutter actually decided
In Thaler v. Perlmutter, the DC Circuit affirmed the Copyright Office's refusal to register an artwork that Stephen Thaler had identified as autonomously generated by his “Creativity Machine.” The court held — narrowly but firmly — that the Copyright Act's requirement of an “author” means a human author. Without a human in the chain of creation, no copyright vests.
The holding sounds sweeping but is in fact quite narrow. Thaler had explicitly disclaimed his own authorship, characterizing the work as the autonomous output of a machine. The court did not address — and was not asked to address — whether a work that involved meaningful human creative contribution but used AI as a tool could be registered. That open question is where most of the practical action now sits.
Two threshold consequences flow directly from the holding. First, models and their operators cannot be authors; only humans can. Second, the Copyright Office's longstanding position that human authorship is a registration prerequisite is now backed by appellate authority. Other circuits considering the question are likely to follow.
II · Office guidanceThe Copyright Office Compendium and current registration practice
The Copyright Office's Compendium of U.S. Copyright Office Practices (third edition, with successive 2023 and 2024 updates) provides the operational rules. The key passages: (a) works produced by “a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author” are not registrable; (b) applicants must disclose AI-generated portions of submitted works; (c) the Office will register works only to the extent of the human contribution.
The disclosure requirement is doing significant practical work. An applicant who fails to disclose AI involvement risks having a registration cancelled. The Office's March 2023 guidance on AI-assisted works walked through several concrete scenarios: an AI-generated image with substantial human selection and arrangement is partially registrable (the human contributions, not the image itself); a work where AI was used to refine human drafts may be fully registrable; a work where AI generated the substantive content based on human prompts is generally not registrable.
The line is being tested case by case. Zarya of the Dawn, a graphic novel produced with Midjourney, became the first high-profile partial-registration outcome — text and arrangement registered, individual AI-generated images not. The Compendium's framework is internally coherent but produces fragmented protection that is awkward to enforce in litigation.
III · The creative sparkHow much human contribution is enough
The doctrinal phrase is “creative spark” — the requirement, dating to Burrow-Giles Lithographic Co. v. Sarony (1884), that copyright protects only original expression that owes its origin to human creative labor. The threshold is famously low. Feist Publications v. Rural Telephone Service requires only a “modicum of creativity.” A grocery list arrangement can qualify if the arrangement reflects choice. The question is whether AI workflows preserve that low threshold or undermine it.
Three patterns are emerging. Prompt-only generation — typing a prompt and using the first output — is consistently treated as insufficient. The prompt is an idea, not protectable expression; the model fills in the expression. Iterative refinement — many prompts, comparing outputs, rejecting and revising — is closer to the line. Human selection among machine outputs has historically been treated as authorship in photography (the photographer's choice of moment, framing, exposure), but that analogy has not yet been definitively extended to AI outputs. Substantial human modification of AI output — painting over, restructuring, recombining with original work — is generally treated as authorship for the human contributions, less so for the underlying machine output.
The practical guidance for creators is to document the chain of creation: prompt history, intermediate outputs, selection rationale, manual edits, and the final composition decisions. The same documentation that protects against false-positive AI accusations also supports a meaningful authorship claim.
IV · InternationalHow other jurisdictions are handling authorship
The US human-authorship rule is one of the stricter positions globally. The United Kingdom has, since 1988, expressly recognized authorship in computer-generated works that lack a human author — assigning it to “the person by whom the arrangements necessary for the creation of the work are undertaken” (CDPA §9(3)). This unusual provision was originally drafted with industrial design in mind but applies cleanly to AI outputs. The UK is therefore the most permissive major jurisdiction for AI-authored works; whether the protection is robust in practice has not been definitively tested.
The European Union requires “the author's own intellectual creation” per the InfoSoc Directive and CJEU case law. The threshold is higher than the US — requiring originality reflecting personal personality — and AI-generated works without meaningful human creative direction are not protected. EU member states implement this similarly.
China has produced perhaps the most interesting comparative jurisprudence: in 2024, the Beijing Internet Court held that an AI-generated image qualified for copyright protection where the human prompter had exercised substantial creative direction over the output. The reasoning was structurally similar to a US human-authorship analysis but reached the opposite practical conclusion. The case is not binding outside China but has been widely cited.
V · Practical guidanceHow to register an AI-assisted work today
For US applicants seeking registration of AI-assisted works, the practical workflow has six steps:
- Document creation chronologically. Save prompt history, intermediate outputs, drafts, and the final piece. Each step that involves human creative decision-making strengthens the application.
- Identify the human-authored elements precisely. Is the human contribution selection? Arrangement? Substantive modification? Original text or imagery created without AI? List specifically.
- Disclaim AI-generated portions. The Copyright Office expects disclosure. Volunteering it produces a cleaner registration than discovery later.
- Apply for registration of the human-authored portions only. Don't claim authorship of AI-generated material. Claim what is yours.
- Consider trade-dress and contract alternatives. Where copyright is uncertain, well-drafted contracts and trademark protection on distinctive elements (titles, characters, marks) provide complementary protection that does not depend on the authorship analysis.
- Reapply if the rules shift. The Office is actively updating guidance. A work that wasn't registrable in 2023 may be registrable today, or vice versa.
SD Frivolous Editorial
The SD Frivolous editorial team combines legal practitioners, journalists, and technologists focused on AI content law. Analysis is peer-reviewed by counsel before publication.
This analysis is journalism and commentary, not legal advice. Laws governing AI content change rapidly. Consult qualified counsel for specific legal questions.